Comparative Study Between ResNet And EfficientNet Family For Classification of Leukemia
Tushar Ibtekar, Md. Rakibul Haque, Azmain Yakin Srizon · 2023
Leukemia, a prevalent hematological malignancy, presents significant challenges in diagnosis and classification. Traditional diagnostic methods, often reliant on microscopic examination, suffer from subjectivity and inconsistency. In an era where timely and accurate diagnosis is critical, identifying the most effective deep learning architecture for leukemia classification becomes imperative. This paper addresses this urgent need by conducting a systematic comparative study between two leading deep learning architectures, ResNet and EfficientNet. Utilizing a generalized method architecture for both models, the study aims to discern which architecture offers a superior balance of accuracy, computational efficiency, and clinical adaptability. EfficientNetB5, the top-performing model in our study, achieved an impressive accuracy of 95.87% and an ROC area of 0.94, underscoring its robustness and reliability in leukemia classification. Performance evaluation metrics, including F1-Score, Recall, Precision, Accuracy, are utilized to conduct a thorough assessment of the effectiveness of these architectural models. By providing a transparent and replicable research process, this study not only contributes to the broader discourse on methodological rigor in deep learning research but also offers a standardized approach for leukemia classification. The findings are intended to guide future research and clinical applications, thereby addressing a critical challenge in healthcare.